Automated Diabetic Retinopathy Detection System

Automated Diabetic Retinopathy Detection System ABSTRACT DETECTION OF EXUDATES USING GUI Automated diabetic retinopathy detection system is an essential requirement due to developing diabetic retinopathy patients around the globe. The primary intention of the research is to detect exudates in digital fundus image for diabetic retinopathy. In this particular study, we provide an efficient method for identifying and classifying the exudates as soft exudates and hard exudates. Apart from these, this study compares three methods namely Contrast Limited Adaptive Histogram Equalization, Histogram Equalization and Mahalanobis Distance for enhancing a digital fundus image to detect and choose the best one to classify exudates in Retinal images by adopting graphical user interface with the help of MATLAB. From the findings of the study, in the image enhancement application of blood vessels, Mahalanobis distance is recognized as the best algorithm. It was evident from the analysis that the monitoring and detecting exudates in the f undus of the eye are essential for diabetic patients. Moreover, it shows that hard and soft exudates are a primary tool of diabetic retinopathy that can be quantified automatically. In addition to these, it appears that drawbacks must be resolved to predict an appropriate detection method for exudates in digital fundus images. From the findings, it was evident that suitable algorithm has to be selected and verified on several images which provide likely and excellent outcomes. LIST OF TABLES Comparison of Histogram Equalization (HE), Contrast Limited Adaptive Histogram Equalization (CLAHE) and Mahalanobis Distance(MD)†¦Ã¢â‚¬ ¦14 LIST OF FIGURE Image before enhancement Histogram before enhancement Image after histogram equalization Histogram after HE Image after CLAHE Histogram after CLAHE Image after Mahalanobis distance enhancement Histogram after Mahalanobis distance enhancement Flow chart of the method CIELab color space Input image K- means clustered image Morphological image Dilated image Eroded image Optic disc detection Exudates image   Hard and soft exudates Input DFI   Enhancement methods of DFI Step-1 of exudate detection Step-2 by giving input image Step-3 enhancing input image Step-4 exudates image of abnormal eye Normal eye output displaying no exudates LIST OF ABBREVIATIONS AHE Adaptive Histogram Equalization CIE Commission Internationale de l’Eclairage CLAHE Contrast Limited Adaptive Histogram Equalization CMYK Cyan, Magenta, Yellow, Key DRD iabetic Retinopathy DFI Digital Fundus Image HE Histogram Equalization MD Mahalanobis Distance MM Mathematical Morphology RGB Red, Green, Blue RRGS Recursive Region Growing Segmentation Chapter 1 Introduction Research Background: Diabetic retinopathy is a common disease nowadays that can prevail in anyone having type 1 or type-2 diabetes. The opportunity of being influenced by this d isease relies on the time duration of a person having diabetes. Long-term diabetes leads to greater blood sugar level that causes harm by changing the flow of blood in retinal blood vessels. It is similar that in the previous stage DR shows no symptoms and hence without facing medical investigation it is not feasible to predict the existence of the disease. Exudative retinopathy is a condition referred by the occurrence of yellow or white mass that exists due to leakage of proteins and fats along with water from vessels of blood in the retina. It is important to predict the exudates occurrence in fundus oculi because the collection of these exudates may lead to complete loss of vision (Manpreetkaur, 2015). Walter et al. (2001) has mentioned that the disease of DR evolved exudates in eye fundus. The physicians regard exudates as one of the primary indicators of DR severity. Exudates are yellow spot resided in fundus. This disease of diabetes causes leakage of fluid from vessels of bl ood. For a long time, uncontrolled diabetes may evolve as exudates in eye fundus. The exudates initiate to develop in little number and size. If the diabetes is not monitored or controlled for a long time the number and size of exudates will grow. The exudates growth in eye fundus may cause blindness. Tasman and Jaeger (2001) have stated that exudates seem as bright deposits of yellow-white on the retina due to lipid leakage from abnormal vessels. Their size and shape differ with various stages of retinopathy. These lesions are related to many diseases of retinal vascular involving DME (diabetic macular edema), DR (diabetic retinopathy), retinal venous obstruction, hypertensive retinopathy, radiation retinopathy and retinal arterial microaneurysms, capillary hemangioma of retina and disease of the coat. Welfera et al. (2010) have stated that exudation is a hazardous case because it can lead to a loss of vision when existing in the central macular area. Thus such lesions must be pred icted, and appropriate medical intervention must be acquired to avoid damages to visual acuity of the patient. Automatic exudates detection in DR patients’ retinas could enhance early prediction of DR and could support doctors track the treatment progress over time. Thus it can be inferred that exudates detection by computer could provide a precise and rapid diagnosis to specialist examination and support the clinician to acquire timely decision to take proper treatment. Problem Statement: Diabetes is a rapidly developing common disease among people globally which causes various organs dysfunction. Diabetic retinopathy is the primary blindness cause in adults. Sometimes, due to long-term diabetes, the retinal blood vessels are harmed, this eye disease is known as diabetic retinopathy. It is essential to automatically predict the lesions of diabetic retinopathy at an early stage to hinder further loss of vision. Exudates are significant diabetic retinopathy symptoms. Exu dates are bright lesions that are an important sign of this disease. It is the major signs of DR a major vision loss cause in diabetic patients. Primary concern of the research Aim: The primary goal of the study is to analyze an automated way for exudates in eyes. Objectives: To examine the causes of exudates in diabetic retinopathy patients. To analyze the types of exudates used in digital fund images. To evaluate the different enhancement methods used to predict the exudates in fundus images. To determine the drawbacks of enhancement methods of exudates in digital fundus images. To propose a promising algorithm to detect the exudates in digital fundus images. Limitations of the study: This study is limited to diabetic retinopathy patients. This study is restricted to exudates detection only. This study evaluates an automated way for exudates in eyes. The structure of the thesis This argument is made up of the following five chapters: Chapter 1: This is the introduction section that gives the necessary research background andconcepts related to the research. Chapter 2 : This chapter is the review of literature that analyzes several existing worksrelated to finding an automated way for exudates in eyes. Chapter 3: This chapter describes the design of the system that explains in detailabout the enhancement methods applied in digital fundus image for detection of diabetic retinopathy. Chapter 4: This chapter discusses the implementation plan of digital fundus images and compares different researches done by authors and depicts the results of the proposed system. Chapter 5: This is the conclusion section that gives the outcome of the research byanswering the research questions and recommendations for future improvement. In addition to that, this thesis has bibliography containing the sources used in collecting secondary data in the study and an appendix that has tools like questionnaires are utili zed in the gathering primary data for the research. Chapter-2 Literature Review Introduction : This chapter provides an overview on the detection of exudates in digital fundus image for diabetic retinopathy. This chapter discusses in detail about the digital fundus image. In addition to these, this chapter discusses in detail about the classification of exudates in retinal images. Apart from these, this study provides the comparison of Histogram equalization (HE), contrast limited adaptive histogram equalization (CLAHE) and Mahalanobis distance (MD) methods to enhance the digital fundus image for detection. Literature on Digital fundus images The benefits of digital imaging are rate of access to information (images), quick and accurate duplication, chronicling and transmission, and prompt access to the outcomes. The imaging technique can be rehashed if the nature of the underlying result is deficient. Despite the fact that film-based images can be digitized (to registe r macular color thickness conveyance from two different wavelength-based pictures or to evaluate the status of the optic nerve), quick access to the images is unrealistic, as it is important to build up the film first. This deferral keeps the picture from checking the outcomes and in this manner redressing any issue in the procurement procedure, which can be efficiently accomplished in digital imaging at no extra cost. The digitization of fundus photos was tended to by (Cideciyan et al., 1991) who proposed a nonlinear rebuilding model fusing four parts: the eye, the fundus camera, the film and the scanner. Scholl et al. (2004) observed digitized images to be valuable for evaluating age-connected maculopathy and age-connected macular degeneration. Comparison Table 1: Comparison of Histogram Equalization (HE), Contrast Limited Adaptive Histogram Equalization (CLAHE) and Mahalanobis Distance (MD) Histogram equalization Contrast limited adaptive histogram equalizatio n Mahalanobis distance This technique is based on the specification of the histogram. CLAHE is considered as the necessary preprocessing step, and it has the tendency to generate the images for extracting the features of a pixel in the classification process. This method has carried out by identifying the pixels of the background images only by leaving the foreground images. HE is relatively straightforward technique and an invertible operator. Indiscrimination is one of the biggest disadvantages of this method. CLAHE is also denoted as the automatic and efficient method to detect the exudates effectively. The selective enhancement of MD has created the fewer artifacts for further processing than HE and CLAHE. HE has used the neighborhood-based approach on the pixels, and it has the tendency to operate based on the modification of histogram to obtain the new images efficiently. The technique of CLAHE has the capabilit y to provide the green channel image enhancement with high quality. This method can produce the similar curve to the Gaussian-shaped curve ideally. HE has uniformly distributed the output histogram by using the cumulated histogram like the mapping function. CLAHE has limited the process of amplification by clipping the histogram at the predefined value. MD algorithm has given better histogram result when compared to HE and CLAHE Research gap : This study examines about the detection of exudates in digital fundus image for diabetic retinopathy. The research gap predicted in this study is that there are many studies on the detection of exudates in digital fundus image for diabetic retinopathy. But no studies have clearly determined the successful approaches towards the detection of diabetic retinopathy in fundus images. Detection and classification of diabetic retinopathy pathologies in fundus images have been investigated by Agurto (2012). He s tudied the effects of image compression and degradation on an automatic diabetic retinopathy screening algorithm. In addition to these, the Agurto et al. (2012) investigated the detection of hard exudates and red lesions in the macula using the multi-scale approach. Walter et al. (2002) carried out an investigation to contribute the image processing to the diagnosis of diabetic retinopathy. Authors also focused on automatic detection of diabetic retinopathy from eye fundus images (Manpreetkaur, 2015). There are also studies that are focused on coarse-to-fine strategy for automatically identifying exudates in color eye fundus images. Chapter-3 Research Design Introduction: This part examines the design of the study to determine an automated way for finding exudates in eyes. This study compares three methods namely CLAHE (Contrast Limited Adaptive Histogram Equalization), Histogram Equalization (HE) and Mahalanobis Distance (MD) for enhancing a digital fundus image to detect and choose the best one to classify exudates in Retinal images by adopting graphical user interface in MATLAB. Research design: The reason of the study is to detect exudates in digital fundus image for diabetic retinopathy. In this particular study, we provide an efficient method for identifying and classifying the exudates as soft exudates and hard exudates. The retinal image seen in the CIELab space of the color is pre-processed for eliminating noise. Further, a network of blood vessels is removed for facilitating detection and removing the optic disc. At the same time, optic disc is removed using the technique of Hough transform. Candidate exudates are identified using the method of k-means clustering. At last, exudates are categorized as the soft and hard one by their threshold and edge energy. Developed method has yielded better outcomes. Histogram Equalization: Histogram equalization is a technique for adjusting image intensities to enhance contrast. HE is an operatio n that is based on histogram specification or modification to obtain new pictures. The objective of this contrast enhancement technique is to get a new enhanced image that has a uniform histogram that only plots the frequency at each gray-level from 0 (black) to 255 (white). Each histogram represents the frequency of occurrence of all gray-level in the image. Figure 1: Image before enhancement Figure 2: Histogram before equalization Figure 3: Image after histogram equalization Figure 4: Histogram after histogram equalization Contrast Limited Adaptive Histogram Equalization: CLAHE is considered as a locally adaptive method for contrast enhancement. CLAHE is an enhanced version of adaptive HE (AHE) method. The technique AHE has a realistic restriction that homogenous part in the image leads to over-amplification of noise due to thin series of pixels are plotted to a whole range of visualization. In the meantime, it was noticed that contrast limited AHE (CLAHE) was designed for preventing this noise over-amplification in homogenous regions. CLAHE restricts the sound amplification in the image in such a way that image looks like very real. Figure 5: Image after CLAHE Figure 6: Histogram after CLAHE Mahalanobis Distance: Image enhancement using the Mahalanobis distance method is performed by identifying the background image pixels and eliminating them, leaving only the foreground image. It is based on the assumption that in image neighborhood N, the background pixels has significantly different intensity value than those of the foreground pixels. For each pixel (x, y) in the picture, the mean  µn (x, y) and the standard deviation à Ã†â€™n (x, y) of the statistical distribution of intensities in N are estimated. The sample means;  µn is used as the estimator for  µn (x, y) and the e sample standard deviation; à Ã†â€™ n is the estimator for à Ã†â€™n (x, y). If the intensity of pixel (x, y) is close to the mea n intensity in N, it is considered to belong to the background set ÃŽÂ ². As defined mathematically in Eq. 1, the expression implies that pixel (x, y) belongs to ÃŽÂ ² if the stated condition is satisfied. Those images would later be combined to evaluate the MD image, which can be segmented using the threshold t to identify the background pixels. Figure 7: Image after MD enhancement Figure 8: Histogram after MD enhancement Summary: This research compares three methods namely CLAHE, HE, MD to enhance a digital fundus image to detect and choose the best one to classify exudates in Retinal images by adopting graphical user interface in MATLAB. It was evident from the above findings that candidate exudates are identified using the technique of Mahalanobis Distance enhancement.